Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs

Jacob Carse (Lead / Corresponding author), Stephen McKenna

Research output: Chapter in Book/Report/Conference proceedingConference contribution

7 Downloads (Pure)

Abstract

Methods to reduce the need for costly data annotations become increasingly important as deep learning gains popularity in medical image analysis and digital pathology. Active learning is an appealing approach that can reduce the amount of annotated data needed to train machine learning models but traditional active learning strategies do not always work well with deep learning. In patch-based machine learning systems, active learning methods typically request annotations for small individual patches which can be tedious and costly for the annotator who needs to rely on visual context for the patches. We propose an active learning framework that selects regions for annotation that are built up of several patches, which should increase annotation throughput. The framework was evaluated with several query strategies on the task of nuclei classification. Convolutional neural networks were trained on small patches, each containing a single nucleus. Traditional query strategies performed worse than random sampling. A K-centre sampling strategy showed a modest gain. Further investigation is needed in order to achieve significant performance gains using deep active learning for this task.
Original languageEnglish
Title of host publicationDigital Pathology
Subtitle of host publication15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings
EditorsConstantino Carlos Reyes-Aldasoro, Andrew Janowczyk, Mitko Veta, Peter Bankhead, Korsuk Sirinukunwattana
Place of PublicationSwitzerland
PublisherSpringer
Pages20-27
Number of pages8
ISBN (Electronic)9783030239374
ISBN (Print)9783030239374, 9783030239367
DOIs
Publication statusPublished - 2019
Event15th European Congress on Digital Pathology (ECDP) - University of Warwick, Warwick, United Kingdom
Duration: 10 Apr 201913 Apr 2019
Conference number: 15th
https://www.ecdp2019.org/

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11435 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th European Congress on Digital Pathology (ECDP)
Abbreviated titleECDP 2019
CountryUnited Kingdom
CityWarwick
Period10/04/1913/04/19
Internet address

Fingerprint

Pathology
Neural networks
Learning systems
Costs
Sampling
Image analysis
Throughput
Problem-Based Learning
Deep learning

Keywords

  • Active learning
  • Deep learning
  • Image annotation
  • Nuclei classification

Cite this

Carse, J., & McKenna, S. (2019). Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs. In C. C. Reyes-Aldasoro, A. Janowczyk, M. Veta, P. Bankhead, & K. Sirinukunwattana (Eds.), Digital Pathology: 15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings (pp. 20-27). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11435 LNCS). Switzerland: Springer . https://doi.org/10.1007/978-3-030-23937-4_3
Carse, Jacob ; McKenna, Stephen. / Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs. Digital Pathology: 15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings. editor / Constantino Carlos Reyes-Aldasoro ; Andrew Janowczyk ; Mitko Veta ; Peter Bankhead ; Korsuk Sirinukunwattana. Switzerland : Springer , 2019. pp. 20-27 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{bcc1d4df4be04305bf2c7b779cd3b751,
title = "Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs",
abstract = "Methods to reduce the need for costly data annotations become increasingly important as deep learning gains popularity in medical image analysis and digital pathology. Active learning is an appealing approach that can reduce the amount of annotated data needed to train machine learning models but traditional active learning strategies do not always work well with deep learning. In patch-based machine learning systems, active learning methods typically request annotations for small individual patches which can be tedious and costly for the annotator who needs to rely on visual context for the patches. We propose an active learning framework that selects regions for annotation that are built up of several patches, which should increase annotation throughput. The framework was evaluated with several query strategies on the task of nuclei classification. Convolutional neural networks were trained on small patches, each containing a single nucleus. Traditional query strategies performed worse than random sampling. A K-centre sampling strategy showed a modest gain. Further investigation is needed in order to achieve significant performance gains using deep active learning for this task.",
keywords = "Active learning, Deep learning, Image annotation, Nuclei classification",
author = "Jacob Carse and Stephen McKenna",
note = "The proceedings of the technical papers will be indexed and published in the form of an edited book, publisher to be confirmed. A selection of papers will be invited to submit extended versions to a special issue in a high-impact ISI indexed journal. https://warwick.ac.uk/fac/sci/dcs/research/tia/ecdp2019/",
year = "2019",
doi = "10.1007/978-3-030-23937-4_3",
language = "English",
isbn = "9783030239374",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer",
pages = "20--27",
editor = "Reyes-Aldasoro, {Constantino Carlos} and Andrew Janowczyk and Mitko Veta and Peter Bankhead and Korsuk Sirinukunwattana",
booktitle = "Digital Pathology",

}

Carse, J & McKenna, S 2019, Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs. in CC Reyes-Aldasoro, A Janowczyk, M Veta, P Bankhead & K Sirinukunwattana (eds), Digital Pathology: 15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11435 LNCS, Springer , Switzerland, pp. 20-27, 15th European Congress on Digital Pathology (ECDP), Warwick, United Kingdom, 10/04/19. https://doi.org/10.1007/978-3-030-23937-4_3

Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs. / Carse, Jacob (Lead / Corresponding author); McKenna, Stephen.

Digital Pathology: 15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings. ed. / Constantino Carlos Reyes-Aldasoro; Andrew Janowczyk; Mitko Veta; Peter Bankhead; Korsuk Sirinukunwattana. Switzerland : Springer , 2019. p. 20-27 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11435 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs

AU - Carse, Jacob

AU - McKenna, Stephen

N1 - The proceedings of the technical papers will be indexed and published in the form of an edited book, publisher to be confirmed. A selection of papers will be invited to submit extended versions to a special issue in a high-impact ISI indexed journal. https://warwick.ac.uk/fac/sci/dcs/research/tia/ecdp2019/

PY - 2019

Y1 - 2019

N2 - Methods to reduce the need for costly data annotations become increasingly important as deep learning gains popularity in medical image analysis and digital pathology. Active learning is an appealing approach that can reduce the amount of annotated data needed to train machine learning models but traditional active learning strategies do not always work well with deep learning. In patch-based machine learning systems, active learning methods typically request annotations for small individual patches which can be tedious and costly for the annotator who needs to rely on visual context for the patches. We propose an active learning framework that selects regions for annotation that are built up of several patches, which should increase annotation throughput. The framework was evaluated with several query strategies on the task of nuclei classification. Convolutional neural networks were trained on small patches, each containing a single nucleus. Traditional query strategies performed worse than random sampling. A K-centre sampling strategy showed a modest gain. Further investigation is needed in order to achieve significant performance gains using deep active learning for this task.

AB - Methods to reduce the need for costly data annotations become increasingly important as deep learning gains popularity in medical image analysis and digital pathology. Active learning is an appealing approach that can reduce the amount of annotated data needed to train machine learning models but traditional active learning strategies do not always work well with deep learning. In patch-based machine learning systems, active learning methods typically request annotations for small individual patches which can be tedious and costly for the annotator who needs to rely on visual context for the patches. We propose an active learning framework that selects regions for annotation that are built up of several patches, which should increase annotation throughput. The framework was evaluated with several query strategies on the task of nuclei classification. Convolutional neural networks were trained on small patches, each containing a single nucleus. Traditional query strategies performed worse than random sampling. A K-centre sampling strategy showed a modest gain. Further investigation is needed in order to achieve significant performance gains using deep active learning for this task.

KW - Active learning

KW - Deep learning

KW - Image annotation

KW - Nuclei classification

UR - http://www.scopus.com/inward/record.url?scp=85069233762&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-23937-4_3

DO - 10.1007/978-3-030-23937-4_3

M3 - Conference contribution

SN - 9783030239374

SN - 9783030239367

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 20

EP - 27

BT - Digital Pathology

A2 - Reyes-Aldasoro, Constantino Carlos

A2 - Janowczyk, Andrew

A2 - Veta, Mitko

A2 - Bankhead, Peter

A2 - Sirinukunwattana, Korsuk

PB - Springer

CY - Switzerland

ER -

Carse J, McKenna S. Active Learning for Patch-Based Digital Pathology using Convolutional Neural Networks to Reduce Annotation Costs. In Reyes-Aldasoro CC, Janowczyk A, Veta M, Bankhead P, Sirinukunwattana K, editors, Digital Pathology: 15th European Congress, ECDP 2019, Warwick, UK, April 10–13, 2019, Proceedings. Switzerland: Springer . 2019. p. 20-27. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-23937-4_3